voiage.experimental_design
Lightweight interfaces for experimental VOI workflows.
Bayesian model fitting and neural amortization remain optional backend responsibilities, so importing this module never requires JAX or NumPyro.
Classes
Section titled “Classes”- InformationGainEstimate Monte Carlo estimate and uncertainty for expected information gain.
- ExperimentalDesign A named candidate design with cost and estimated information gain.
Functions
Section titled “Functions”- expected_information_gain Estimate expected information gain from simulated log ratios.
- select_bayesian_design Select the candidate with greatest information gain per unit cost.
- select_active_learning_batch Return indices of the highest-scoring candidates for active learning.
- amortized_evsi Summarize simulated amortized EVSI predictions against current care.